AVR Freaks

Well yes, I read them when programming in serial mode. My programming utility compares them to a table to decide what mode of programming to use, and to verify that it is the right sort of chip for the code compiled. I can force it to assume a particular chip type regardless of sig bytes read, but then I lose the self-checking ability.

Slightly more concerning is that the calibration value for the internal RC oscillator is held in the signature byte area, so my chip will presumably calibrate the oscillator at FF, which may well be too fast for reliable EEPROM writing. I assume this calibration byte is written post-manufacture, which suggests that the whole signature area is likely to be writable - and probably eraseable, which is what I appear to have done.

I suspect there are undocumented codes in the serial programming system which can write to these bytes, but rather than try codes at random (and risk damaging something else), I wondered whether anyone here already knew what they were.

The Mega16 has a 0x40-byte signature area, and this has all been erased to FF. The OSCCAL byte is the most-significant half of the first word of the sig area, with the identifying sig bytes in the least-significant halves of words 0-2. Some chips have non-FF values in other bytes of the sig area, but I haven't figured out what any of these mean.

Since the only thing that happened between programming correctly and losing sig bytes was that I ran a program compiled for the wrong address space, I assume something that program did managed to erase the sig area. This suggests it's either related to normal EEPROM writes, or possibly self-programming (though I would have thought it was difficult to get into self-programming mode accidentally).